Open Safety Institute

Attackers are becoming agents.

Autonomous AI systems can discover vulnerabilities, exploit software, and adapt while an intrusion is still underway.

Defenders need to see what an attacker agent is doing as it happens, contain it before it can keep moving, and understand how to protect the systems it targeted.

We build the open evaluations, detections, and defensive tools they need.

Our focus is active attacker agents.

We study how they operate, identify the traces they leave behind, and turn that evidence into tools defenders can test and use.

We publish our work openly so the defensive advantage is shared.